Module 03 Activity

Scenario

Classical processing is fast, free and inspectable. This activity finds where your pipeline's hidden assumptions are.

What you build

A processing pipeline tested across the real range of conditions, with every hard-coded constant identified.

Steps

1. Sweep the blur kernel and measure both noise reduction and edge softening at each setting. Choose one and justify it from your feature size.
2. Apply sharpening after denoising and after not denoising. Measure the flat-area noise in both.
3. Compare a global threshold against an adaptive one on an unevenly lit image. Report where each fails.
4. Apply opening then closing, and then the reverse order. Report the component count for both.
5. Collect images from at least three real conditions and run your fixed pipeline on all of them.
6. List every hard-coded constant in the pipeline and say what it assumes.

Evidence to hand in

• The blur sweep with both measurements and your chosen kernel.
• The sharpening comparison.
• Global versus adaptive on uneven lighting.
• Morphology in both orders with component counts.
• Pipeline results across three real conditions.
• The hard-coded constant audit.

Review checklist

• The kernel choice is justified by feature size, not by appearance.
• The morphology order is opening then closing, and the reverse is shown to be worse.
• At least three genuinely different conditions were tested.
• Every hard-coded constant is listed with its assumption.
• A failure on one condition is reported rather than tuned away.

Mark complete